A preliminary investigation towards improving linked data quality using distance-based outlier detection

Abstract

With more and more data being published on the Web as Linked Data, Web Data quality is becoming increasingly important. While quite some work has been done with regard to quality assessment of Linked Data, only few works have addressed quality improvement. In this article, we present a preliminary an approach for identifying potentially incorrect RDF statements using distance-based outlier detection. Our method follows a three stage approach, which automates the whole process of finding potentially incorrect statements for a certain property. Our preliminary evaluation shows that a high precision is maintained with different settings.